Kyle Min

Error with tensorflow 1.2.1

1

If I try to train the model on Visual Genome, this error occurs. (tensorflow 1.2.1) Traceback (most recent call last): File "./exp-visgeno-rel/exp_train_visgeno_attbilstm_strong.py", line 65, in text_seq_batch, num_vocab, embed_dim, lstm_dim, vgg_dropout,...

Cannot reproduce the reported results of ViG when using the pretrained weight files

5

I tried to reproduce the results of ViG with the released weight files as seen below. However, what I obtained are top1: 79.70% (top5: 95.07%) for ViG-S. I tried multiple...

Interpolation of pos_embed and relative_pos for larger input

2

Hi, I wonder if I can use the pre-trained model to process larger images (e.g. 256x256 or 384x384). I tried changing `self.pos_embed` and `HW` and loading the pre-trained parameters of...

update for CVPR and ECCV 2020 on both supervision (+typo corrections)

Cannot reproduce the reported performance

3

The reported performance cannot be reproduced unless I use the pre-trained model. If I train the model from scratch, the performance never reaches the reported ones. Can you share random...

I used exact same image preprocessing, including cropping and resizing. However, as training goes on, g_loss increases gradually, and the training and test image look very strange. (no matter what...

does not work with EOS M 100 (pycraw2_demo.py)

3

I tried to convert CR2 files taken from EOS M 100 by using pycraw2_demo.py. However, all the converted tiff files do not have the correct color space. I think that...